Microseismic monitoring systems enable real-time acquisition of vibration signals generated by coal and rock fracturing, deriving key parameters (e.g., event time, spatial location, and energy release) via inversion algorithms for rock burst risk prediction in coal mining operations. However, the underground coal mining environment is highly complex, and signals acquired by sensors are often contaminated with substantial blasting-induced interference, thereby significantly degrading the accuracy of monitoring and early warning. While existing research predominantly utilizes time-frequency transform and fractal analysis to characterize mining microseismic and blasting signals, systematic investigations into data-driven automatic identification of these two signal types remain conspicuously limited. To address this gap, this study investigated the characteristic differences between mining microseismic signals and blasting signals across three analytical dimensions: time-domain, frequency-domain, and fractal dimension. Using the Relief characteristic selection algorithm, discriminative characteristics for distinguishing the two signal types were selected. Moreover, discrimination models for mining microseismic and blasting signals were developed using three machine learning algorithms: Fisher's linear discriminant analysis, support vector machine (SVM), and adaptive boosting (AdaBoost). The results indicated that the discriminative characteristics for distinguishing mining microseismic signals from blasting signals include duration, rise time, post-peak attenuation coefficient, high-frequency/low-frequency ratio, and multifractal spectrum width (Δα). In terms of model performance, the classification model based on AdaBoost exhibited good stability, whereas the SVM model achieved the highest identification performance with an overall recognition accuracy of 91.2%. These models can effectively satisfy the real-time monitoring requirements of engineering sites.